Why Develop Demand Planning
Factory Demand Planning
- Lack of high quality business data
- Single data source, fragmented management, unable to build effective data correlation
- High data noise, which affects forecasting methods and results
- Lack of in-depth data mining
Operation challenges
- Variety of SKUs is getting larger and larger, with a large number of long-tail goods,
resulting in a heavy and difficult management burden
- Growing demand volatility and increased complexity due to external factors:
- Promotions
- Competitor promotions
- Holidays, large events, weather, etc.
Sales opportunities go fast
- Identifying missed sales opportunities due to stock out
- Reducing the impact of stock-outs in demand forecasting
- How to ensure proper replenishment of goods to ensure：
- Best sellers are not restocked
- No backlog of general merchandise inventory
Manual decision making challenges
- Inability to utilize multi-dimensional data
- Decision quality fluctuates greatly and is difficult to quantify
- High employee turnover, experience of good employees cannot be accumulated
DP - A smarter way to forecast demand
Artificial intelligence can not only consider far more dimensions and factors than individuals and
teams, but also can continuously learn and iterate
Predictive Process Model
1
Data import and governance
Intelligent identification of abnormal and characteristic data based on statistics and big data analysis
technology, automatic de-duplication, rejection, interpolation and smoothing of outliers, identification
of restricted sales periods.
Historical sales orders
geographic information of business areas
Product life cycle
consumer portrait
category characteristics
Weather and climate
new product launch plan
holiday calendar
promotional events
abnormal sales
2
Feature construction and slicing analysis
After inputting of the governed data, we automatically analyze which influence factors are positively
associated with sales, and intelligently build feature quantities through machine learning, deep
learning, neural networks and other artificial intelligence technologies.
3
Multi-level hybrid model prediction and fusion
Comprehensive use of time series, machine learning, deep learning and other frontier algorithms in
various fields, iterative training sample data, for different types of goods, capable learn the most
appropriate model fusion.
4
Prediction bias correction and model Performance Optimization
Combining with various built-in algorithms, the system intelligently calculates the forecast accuracy
and forecast deviation rate, and obtains the model with the smallest deviation and corresponding
parameter settings for a specific cycle and dimension.
Get the latest sales data
Forecast accuracy
Bias MAPE
Forecast correction
T+1
Forecast monitoring
Continuous Model Performance Optimization
Product Strengths
UHAlean's demand planning product organically blends algorithms and human experience in a process that
improves forecast accuracy by leveraging the strengths of each.
Algorithm Strengths
-
Mining laws: Machine learning, data mining and other algorithms can effectively mine the
hidden laws behind high-dimensional data and identify the combination of factors with a
significant impact on sales
- Automatic adaptation: automatically adapting to the sales patterns of different products,
channels and time periods
Human experience advantage
- Flexible response: mastering more non-standard information, such as local emergencies, etc
- Manual Identification: better detection of data noise
Forecast Product Value
Value that demand predicting products and solutions can bring to customers
Forecast accuracy improvement
In comparison to traditional forecasting methods, the prediction accuracy of core products is
expected to improve by 13%, and the prediction accuracy of all products is expected to improve
by 11%.
Forecast labor input saving
Through the automatic parameter adjustment function of the model, the manual input of the
forecasting department is reduced, and the prediction cost is significantly decreased
Forecast stability improvement
Compared with traditional manual prediction, Machine Learning algorithm forecasts provide a more
stable effect
Forecast dimension customization
Time dimension can be customized to better fit the actual use case
